The usual way into a national holiday post is to open the template gallery, find the one everyone is using, and drop a selfie into it. On 10 September 2026 that search came back with Teacher's Day and Thanksgiving Day, and the local-language spelling of the holiday returned nothing at all. So this piece builds the portrait from a written request instead, at 16:9, from one generated selfie, and measures the three things that decide whether the result is usable: the face, the backdrop, and the hands.
The finished portrait, generated at Seedream 4.5, 16:9 and 2K from the selfie further down this page. The person is invented; no photograph of a real person was used.
The template search that comes up empty
The template explorer, Image tab, on 10 September 2026. Both search terms are highlighted; the results are what the gallery returned on this account that day.
The English search returns 109 items and they are templates with the word day in the name. The holiday in its own language returns zero. That is worth two minutes of your own checking before you plan a post around a template, because the gallery changes and your market may not be this one, but it also sets up the rest of this article: everything below is written out rather than picked off a shelf.
The gallery is at template explorer in the workspace, with a Video and an Image tab. The portrait work happens in image mode through the AI image generator flow.
The photograph you start from
The source photograph, generated for this article at Seedream 4.5, 16:9 and 2K. The face box is 200 pixels wide when the frame is 1280 wide.
Photoreal photograph, 16:9 landscape. A woman in her late twenties looks straight at the camera from the middle of the frame, head and shoulders only, standing in front of a plain light gray wall, wearing a plain dark gray T-shirt, hair tied back, no jewelry. Even soft light from the front, no hard shadow on the wall, sharp focus on her eyes. Space above her head and both shoulders inside the frame. No text, no logos, no brand marks.
Three things about that source matter later. The light is flat and frontal, so nothing in the next step has to invent a shadow. The head is large in the frame, which is what gives the checks below something to work with. And even though the request said the middle of the frame, she came back to the right of it with a wide stretch of empty wall on the left, which is what a 16:9 frame does to a single upright subject unless you crop it yourself.
If you use your own photograph instead of a generated one, the same three apply: front on, one soft light, head large enough to fill a good part of the height.
What a one-line request decides for you
The obvious first attempt is a sentence: turn this into an Independence Day portrait in red and white. What came back was a different country's holiday. The model filled the background with bunting and a striped field belonging to a market this picture was not made for, and it redrew the plain T-shirt as a mottled red garment with a pattern in it. The face survived; almost nothing else did.
That image is not reproduced here, and the reason is the useful part: a machine-drawn version of any country's flag is not something to publish. Flags and emblems are drawn to a specification, a generator draws them from an impression, and the difference shows. Keep them out of the request and put them in later from a file you control, or leave them out of the picture.
The practical rule that follows: name the colors and name the objects that carry them. Two words of holiday framing hand the whole decision to the model, including which country the picture belongs to.
Writing the picture instead
Keep her face, her hairline and her skin exactly as they are in the photo. Photoreal photograph, 16:9 landscape, head and shoulders, facing the camera, with space above her head. She wears a plain white shirt buttoned at the collar and a plain red jacket over it, both in flat colors with no pattern. Behind her the wall is covered by two plain cloth panels hanging flat and straight, red on the left half and white on the right half, with a straight vertical edge between them. Even soft light from the front, no hard shadow on the wall, sharp focus on her eyes. No flags, no emblems, no stars, no stripes, no text, no logos, no brand marks.
Each clause is doing one job. Keep her face, her hairline and her skin exactly as they are is the identity clause. Plain and flat colors with no pattern are what stopped the mottled fabric. Two plain cloth panels replaces the flag with an object that is only color. And the closing list is a fence: no flags, no emblems, no stars, no stripes, no text.
To move this to another market, change the two color words and leave the rest of the paragraph alone. That is the part of the page that is worth reusing.
Is it still the same face
Faces cropped from the eyes to the chin at full resolution and compared with the photograph. The numbers are matching points that survive a geometry check.
The comparison works by finding small features in each face, matching them between two pictures, and keeping only the matches that fit one consistent geometry. The photograph against itself gives 229, which is the ceiling. A second person generated from the same selfie prompt gives 18, which is the floor. The written-out portrait scores 51 and the hands version 59, both far above that floor: the face carried over.
Two things in those numbers are worth carrying into your own checks. The versions match each other more closely than any of them matches the photograph, at 75 to 87 points: they drift together, in the same direction, rather than each going its own way. And the waist-up version scored 28, close to the floor, because its face is 300 pixels wide instead of 400 and there is less of it to compare. The wider you frame, the less of the identity you can check afterward.
Read by eye at the same size, the drift shows in three places on these images: the lips are fuller and more saturated than in the photograph, the eyebrows are heavier, and the skin is smoother. Nothing about the spacing moved. The distance between the eyes as a share of face width is 0.379 in the photograph and 0.369 to 0.391 across the versions, which is the same face at the same proportions with a retouched surface on it.
Where the divide actually lands
The request asks for red on the left half and white on the right half. Measured on the export, the divide sits at 41.3 percent of the frame width in the first run, 49.7 in the second, 40.7 in the third.
Three runs of the same sentence. The white line on each thumbnail marks the measured divide; the percentages are of the full frame width.
The edge itself is not the problem. In two of the three runs, 94 percent of the pixel rows sat within five pixels of a single vertical line, so what you get is a clean edge in an unpredictable place. Running the same description with nobody in the picture settles the question of why.
The same backdrop sentence with and without a person. The white line is the measured divide, the green line is the middle of the frame.
With an empty wall the divide came back at 49.8 percent twice, with 99 percent of rows within five pixels of vertical. Put a person in front of it and the same words land near 41. The subject of the picture gets the geometry; the background gets what is left.
So do not build meaning into where that edge falls, and do not promise a client a half-and-half backdrop from a written request. Check it on the file, and crop the frame if the balance matters.
Hands, in the frame or out of it
One clause changed between these two runs: where the hands are and whether they have to be inside the frame.
Hands at waist height fall on the bottom edge of a 16:9 frame when the subject is framed head and shoulders, and half a hand tells you nothing. Raised to chest height, with fully inside the frame, held apart from each other, palms toward the camera and all fingers separate, both hands came back complete on the first try: five fingers each, thumbs on the inside.
The pattern behind that is worth keeping. Hands go wrong where they overlap something, including each other and the body, so a request that separates them and shows them flat is asking for the easy version of the problem. If the pose you want has hands crossing or holding something, plan on more than one attempt.
What the ten requests cost
The cost lines as CapCut showed them under each request, read on 10 September 2026.
Ten credits, of which four were spent on comparisons rather than on the picture: a second person to give the face check a floor, an empty backdrop to explain the divide, and two runs of a backdrop test that was dropped. A reader who only wants the portrait needs three of the ten. These figures are for these configurations on this account and this date and come from the line the composer shows before each request; other models, sizes and modes are charged differently, and prices change.
What is not covered here: stored templates, since the gallery had none for this holiday on this date; background removal, local repainting and image enhancement; 4K export; video versions of the same idea; and captions or greeting copy. The portrait here is a generated person, and if you build this from a photograph, use one of yourself or of someone who has agreed to it.
Next year, change these lines
This page comes around on the same date every year, and the parts that expire are small. The two color words in the backdrop clause and the jacket clause carry the market. The year and the anniversary number, if you put them in the caption rather than the picture, sit outside the request entirely, which is the reason not to ask the generator for text. The rest of the paragraph, the identity clause, the flat-color clause, the fence at the end, is what you keep.
Before you post, three checks, in the order they cost you time: open the source photograph and the result at the same size and look at the lips, the eyebrows and the skin; measure or eyeball where the backdrop divide fell and crop if it matters; and count the fingers on any hand you can see. Two of those three go wrong quietly, which is why they are worth a minute before publishing rather than an hour afterward.
The planning row for this page called it a photo template. No template was used, because the gallery search shown above returned none for this holiday on the date of writing, so the title says what the article does instead. The address of the page is unchanged.
Written 10 September 2026. Every image here was generated in CapCut for this article; the two people in it are invented, and no real person, party, official or organization appears. No national flag or emblem was generated, reproduced or altered: the backdrop is two plain cloth panels, and the one request that produced flag imagery is described rather than shown. Interface labels, template search results and cost lines reflect a single session on that date and the configurations named above, and may change. Face comparisons are counts of matching image features that survive a geometry check, not face recognition; backdrop divides were measured column by column on the downloaded files.